Fix mypy errors from convert_to_tensor → as_tensor swap

The new `img.as_tensor() if isinstance(img, MetaTensor) else cast(...)`
expression narrowed mypy's inference of `img_t` to `Tensor`, so downstream
reassignments from `_clip` / `_normalize` / `interp` (which return
`NdarrayOrTensor`) failed type-checking. Annotate `img_t` explicitly as
`NdarrayOrTensor` and drop the redundant `cast(torch.Tensor, img)` so the
variable matches the type used by the rest of each transform's body.

Keep the `cast` only on `SavitzkyGolaySmooth.__call__` where `self.img_t`
is declared as `torch.Tensor` in `__init__` and used unconditionally as a
Tensor afterwards.

Signed-off-by: Soumya Snigdha Kundu <soumya_snigdha.kundu@kcl.ac.uk>

Author: aymuos15

monai/transforms/intensity/array.py

@@ -484,7 +484,7 @@ def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
 
         """
         img = convert_to_tensor(img, track_meta=get_track_meta())
-        img_t = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_t: NdarrayOrTensor = img.as_tensor() if isinstance(img, MetaTensor) else img
         ret: NdarrayOrTensor
         if self.minv is not None or self.maxv is not None:
             if self.channel_wise:
@@ -543,7 +543,7 @@ def __call__(self, img: NdarrayOrTensor, factor=None) -> NdarrayOrTensor:
         factor = factor if factor is not None else self.factor
 
         img = convert_to_tensor(img, track_meta=get_track_meta())
-        img_t = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_t: NdarrayOrTensor = img.as_tensor() if isinstance(img, MetaTensor) else img
         ret: NdarrayOrTensor
         if self.channel_wise:
             out = []
@@ -1169,7 +1169,7 @@ def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
         Apply the transform to `img`.
         """
         img = convert_to_tensor(img, track_meta=get_track_meta())
-        img_t = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_t: NdarrayOrTensor = img.as_tensor() if isinstance(img, MetaTensor) else img
         if self.channel_wise:
             img_t = torch.stack([self._clip(img=d) for d in img_t])  # type: ignore
         else:
@@ -1434,7 +1434,7 @@ def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
         Apply the transform to `img`.
         """
         img = convert_to_tensor(img, track_meta=get_track_meta())
-        img_t = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_t: NdarrayOrTensor = img.as_tensor() if isinstance(img, MetaTensor) else img
         if self.channel_wise:
             img_t = torch.stack([self._normalize(img=d) for d in img_t])  # type: ignore
         else:
@@ -1908,7 +1908,7 @@ def __call__(self, img: NdarrayOrTensor, randomize: bool = True) -> NdarrayOrTen
 
         if self.reference_control_points is None or self.floating_control_points is None:
             raise RuntimeError("please call the `randomize()` function first.")
-        img_t = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_t: NdarrayOrTensor = img.as_tensor() if isinstance(img, MetaTensor) else img
         img_min, img_max = img_t.min(), img_t.max()
         if img_min == img_max:
             warn(
@@ -1953,7 +1953,7 @@ def __init__(self, alpha: float = 0.1) -> None:
 
     def __call__(self, img: NdarrayOrTensor) -> NdarrayOrTensor:
         img = convert_to_tensor(img, track_meta=get_track_meta())
-        img_t = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_t: NdarrayOrTensor = img.as_tensor() if isinstance(img, MetaTensor) else img
         n_dims = len(img_t.shape[1:])
 
         # FT
@@ -2605,7 +2605,7 @@ def __call__(self, img: torch.Tensor) -> torch.Tensor:
             img: image to remap.
         """
         img = convert_to_tensor(img, track_meta=get_track_meta())
-        img_ = img.as_tensor() if isinstance(img, MetaTensor) else cast(torch.Tensor, img)
+        img_ = img.as_tensor() if isinstance(img, MetaTensor) else img
         # sample noise
         vals_to_sample = torch.unique(img_).tolist()
         noise = torch.from_numpy(self.R.choice(vals_to_sample, len(vals_to_sample) - 1 + self.kernel_size))